Knowledge-Directed Automatic Bronchi Detection Method for Sparse HRCT Scans of the Lungs

The identification of bronchi on High-resolution CT (HRCT) images of the lung provides valuable clinical information in patients with suspected airways diseases such as bronchiectasis. This paper presents a new method for automatic detection of bronchi for sparse HRCT data. Using sparse instead of volumetric data helps reduce the amount of data needed to be processed and stored. However, additional effort is required to compensate the information loss between the slice gaps. The detection method is based on local intensity gradient analysis and rule-based classification. The rules are constructed from high-level knowledge about bronchial anatomy. A bronchial tree model is used to guide the detection algorithm about the location and the number of expected bronchi on a particular image. The method was evaluated on 67 HRCT images and achieved 73% sensitivity and 83% precision.